Exploring Graph-structured Passage Representation for Multi-hop Reading Comprehension with Graph Neural Networks
This work addresses the problem of improving answer accuracy in multi-hop reading comprehension for NLP systems, representing an incremental advance over previous methods.
The paper tackled multi-hop reading comprehension by introducing a method that forms complex graphs to better connect global evidence, using graph neural networks for evidence integration. The method outperformed all published results on two standard datasets.
Multi-hop reading comprehension focuses on one type of factoid question, where a system needs to properly integrate multiple pieces of evidence to correctly answer a question. Previous work approximates global evidence with local coreference information, encoding coreference chains with DAG-styled GRU layers within a gated-attention reader. However, coreference is limited in providing information for rich inference. We introduce a new method for better connecting global evidence, which forms more complex graphs compared to DAGs. To perform evidence integration on our graphs, we investigate two recent graph neural networks, namely graph convolutional network (GCN) and graph recurrent network (GRN). Experiments on two standard datasets show that richer global information leads to better answers. Our method performs better than all published results on these datasets.